Fig. 1

Conceptual overview subtype predictors. a Single sample predictor (SSP). For each subtype a centroid is computed (depicted by different colors) representing a vector of average values for each gene in the intrinsic gene list (IGL), i.e. a predetermined list of relevant genes, taken over a training set of samples assumed to be of the same subtype. In order to determine the subtype of a new case, one computes the distance to each of the centroids and assigns the new case to the subtype corresponding to the centroid that is nearest, here assumed to be the luminal A centroid, leading to the luminal A subtype. b Subtype classification model (SCM). Each sample is represented by three module scores (MS) calculated based on module gene lists (MGLs), i.e. the list of genes associated with a module. Training set samples are first divided into basal, HER2 and luminal subtypes by fitting a 3-component Gaussian mixture model to the ER and HER2 related module scores (top panel, colored circles and dotted grey ovals). Subsequently, cases of the luminal subtype are divided into two subtypes, based on their proliferation module score. Samples with a low proliferation score are assigned to the lumA (luminal A) subtype, whereas samples with a high proliferation score are assigned to the lumB (luminal B) subtype. The subtype of a new case can be determined by calculating the posterior membership probabilities under the Gaussian mixture model and selecting the subtype associated with the maximum posterior probability. In the example, the new case (depicted with a cross) has a high ER module score and low HER2 and proliferation module scores, leading to the luminal A subtype. c STG subtype predictor based on the St. Gallen surrogate intrinsic subtype definitions [14]. Over(+)/under(-)expression of clinical markers for ER, HER2, KI-67 (proliferation status) and PGR allows for 2⁴=16 distinct profiles. Here, the over/underexpression status of each marker was determined based on microarray measurements in a way similar to SCMs, i.e. via module scores. The subtype of a new case is fully determined by the over/underexpression status of the individual markers. In the example, the new case is assumed to have a high ER signaling score and low HER2, PGR and proliferation scores, leading to the luminal A subtype (blue arrow)